Codeplay aims to open up developer software stack for new AI accelerators by John Abbott of 451 Research
24 September 2020
Read the report from John Abbott at 451 Research analyzing the opportunity for companies looking to encourage developers to move to their own processors, and how Codeplay can help.
“The availability of new architectures will inevitably challenge NVIDIA’s dominance. Codeplay’s migration capabilities will help open up the choices.”
“Codeplay, a compiler specialist that has worked on GPU-powered videogames for the past two decades, has turned its attention to artificial intelligence (AI). Its goal is to help AI accelerator vendors gain design wins by establishing an open standards-based cross-architecture software layer and ecosystem. Current software is typically tied to a single architecture and often supports only a limited number of graph compilers and programming models. Codeplay offers ‘silicon enablement’ services to customers that include Intel and Renesas. It backs the standard SYCL abstraction layer for the OpenCL framework, designed to execute parallel programming across heterogeneous platforms (CPUs, GPUs, FPGAs, DSPs and AI accelerators). The dominant incumbent in this space is NVIDIA with its CUDA software platform, which has built up a broad and established ecosystem aimed at AI developers. However, CUDA only works on top of NVDIA GPUs.”